Welcome back to LLM Decode 👋
AI is starting to reshape both how intelligence gets built and how we interact with it. Anthropic says Claude is already taking a bigger role inside its own research process, while Meta is reportedly exploring smart glasses that remove one of wearables’ most controversial features: the camera.
Here’s what matters today.
Claude Is Starting to Help Build Its Own Successors
Anthropic says Claude is now deeply embedded in the company’s research and development process. As of August, Claude was leading roughly 26% of Anthropic R&D projects, completing work from high-level human instructions under supervision.
Its involvement is even broader than that number suggests. Anthropic says Claude participates alongside humans in about 90% of its R&D work, while roughly 30,000 AI agents operate across the company’s research environment under oversight mechanisms.
The important distinction: this is not autonomous self-improvement. Humans still set research direction and supervise the systems. Anthropic says it wants standardized measurements and independent evaluation to track how much AI contributes to developing future AI systems.
Why it matters
AI development could accelerate: Models can increasingly assist with experiments, coding and research needed to create future models.
Researchers become orchestrators: More work may shift from executing every task manually to directing and reviewing AI agents.
The feedback loop matters: Better AI → more capable AI research tools → potentially faster development of better AI.
Measurement becomes important: Companies will need ways to distinguish AI-assisted research from genuinely autonomous research.
Meta May Be Building AI Glasses Without the Camera
Meta reportedly plans to launch a new pair of smart glasses, internally called Luna, this fall. The biggest change is surprisingly simple: no camera.
According to reporting on the device, Luna could retain microphones and speakers for voice interaction while removing video capture. The Verge reports that the glasses may include six microphones, work with Meta AI and Muse, and target an October release.
That decision comes as camera-equipped smart glasses face growing privacy scrutiny. Australia is considering restrictions on such glasses in government buildings, while other institutions have already imposed limits on their use.
Why it matters
AI wearables don't necessarily need cameras: Voice-first AI could handle messaging, assistance and information without constantly seeing the environment.
Trust is becoming a product feature: Removing capabilities can sometimes make technology easier to adopt.
AI may move beyond screens: Phones won't disappear, but glasses, earbuds and other ambient interfaces could handle more everyday interactions.
Privacy isn't solved completely: Camera-free glasses can reduce visual-recording concerns, but microphones create their own privacy questions.
4 AI Workflows to Try This Week
AI Research Partner: Question → AI explores → Human reviews → AI tests → Human decides. Instead of using AI only for answers, give it a defined research task and make yourself the reviewer.
Agent Delegation Ladder: Assist → Execute → Verify → Lead. Track which recurring tasks AI can safely move through these stages instead of giving agents full autonomy immediately.
Interface Reduction Test: Pick one workflow and ask, “Could this happen through voice instead of opening five apps?” This helps identify opportunities for voice-first automation.
AI Oversight Log: For important agent workflows, record what AI decided, what tools it used, what humans approved, and the final outcome. As agents do more work, observability becomes as important as capability.
Ready to level up your AI skills?
Explore Our Courses →That’s it for today.
The AI space doesn’t slow down - and neither should your thinking.
See you in the next drop.